An Inhibition/Enhancement network for noise robust ASR

This paper describes an evaluation of Inhibition/Enhancement (In/En) network for noise robust automatic speech recognition (ASR). In articulatory feature based speech recognition using neural network, the In/En network is needed to discriminate whether the articulatory features (AFs) dynamic patterns of trajectories are convex or concave. The network is used to achieve categorical AFs movement by enhancing AFs peak patterns (convex patterns) and inhibiting AFs dip patterns (concave patterns). We have analyzed the effectiveness of the In/En algorithm by incorporating it into a system which consists of three stages: a) Multilayer Neural Networks (MLNs), b) an In/En Network and c) the Gram-Schmidt (GS) algorithm for orthogonalization. From the experiments using Japanese Newspaper Article Sentences (JNAS) database in clean and noisy acoustic environments, it is observed that the In/En network plays a significant role on the improvement of phoneme recognition performance. Moreover, the In/En network reduces the number of mixture components needed in Hidden Markov Models (HMMs).